CONCEPT
Expertise Deference
The gradual erosion of the professional's confidence in her own judgment relative to an AI system she has observed being correct many times—a miscalibration that becomes catastrophic at the override moment, when the AI is wrong and human assertion is the only safety net.
Every effective collaboration requires a calibrated sense of when to defer and when to assert. The surgeon defers to her cardiologist colleague on cardiac questions and asserts her own authority on surgical ones. The pilot defers to the weather service on meteorological questions and asserts her own judgment on approach decisions. These calibrations are built through experience—enough interactions with each source of expertise to develop a reliable internal model of its strengths and limits. The challenge that
Bainbridge's framework identifies in AI-augmented work, and that
Ericsson's framework illuminates at the developmental level, is that this calibration drifts systematically toward deference when the AI is correct in the vast majority of cases. Each correct AI output updates the practitioner's internal model of AI reliability upward. The errors—the outputs that would calibrate trust downward—are precisely the outputs the practitioner is least likely to detect, because her detection capability has been simultaneously degraded